PHAROS: Highly Deformable Proprioceptive Membrane for Real-Time 3D Shape Reconstruction

Demos

Abstract

Reconstructing the three-dimensional (3D) geometry of surfaces is essential for robot perception and physical interaction. Vision-based approaches reconstruct surfaces at high resolution but require a clear line of sight, which is often unavailable at contact interfaces. A conformable membrane that reconstructs its own deformation offers a complementary route. Building on two-dimensional optical waveguide shape sensing, this work presents a stretchable proprioceptive silicone membrane that reconstructs localized indentation and in-plane stretching. The membrane integrates edge-mounted light-emitting diodes (LEDs) and centrally distributed photodiodes (PDs) in a multilayer elastomeric waveguide with liquid-metal interconnects. A neural network maps the time-multiplexed LED–PD intensity measurements to a dense point cloud of the membrane surface. Under localized indentation of up to about 25 mm, the membrane achieves an overall shape reconstruction error of 1.580 mm, and 1.884 mm on an indenter with a novel shape. Under uniaxial stretching of up to 75% strain, the in-plane outline is reconstructed with a mean Chamfer distance of 1.557 mm. Optical acquisition and model inference run in real time at about 90 Hz. These results indicate a practical route to proprioceptive shape sensing in deformable robotic systems.